Defect Detection and Cause Analysis on Display Electrostatic Chuck FAB Process Using Deep Learning Algorithms

Deep learning algorithms have been used in various production processes to reduce product defect rates and improve process management efficiency. In this study, we propose a method for defect detection and cause analysis in the field of display electrostatic chuck fabrication (FAB) processes. The proposed method consists of three steps. First, we construct an autoencoder model for detecting defects and obtain reconstruction errors. Second, we construct the deep neural network model based on reconstruction errors obtained from the preceding autoencoder model to predict defects. Lastly, we applied locally interpretable model-agnostic explanations (LIME) and Shapley additive explanations (SHAP) to the deep neural network model for identifying the major causes of defects. The usefulness and applicability of the proposed method was demonstrated using the actual electrostatic chuck FAB data. Our proposed model outperformed all comparative models by achieving the highest average recall score. Furthermore, lag time and electrostatic chuck driving environment were identified as major causes of the defect.